LoRA, PEFT, instruction tuning and domain adaptation — adapting a pretrained model without paying to train one.
arXiv:2603.16497v3 Announce Type: replace-cross Abstract: Time series foundation models (TSFMs) require diverse, real-world datasets to adapt across varying domains and temporal frequencies. However,...
The paper introduces Influence‑Guided PPO (I‑PPO), a framework that incorporates data attribution into the Proximal Policy Optimization (PPO) post‑training loop for large language models. I‑PPO computes an influence score for each episode via a gradient‑based approximation and removes episodes that are anti‑aligned with a validation gradient, thereby filtering out noisy or unfaithful reasoning. Experiments show that this filtering consistently outperforms both supervised fine‑tuning (SFT) and standard PPO, acting as an intrinsic early‑stopping mechanism that speeds up training and reduces unfaithful chain‑of‑thought reasoning.
PROOF-Gen is a method that improves distillation of tool‑calling models by recovering successful trajectories from teacher failures. It uses per‑scenario prompt optimization to generate corrective guidance that steers the teacher to a passing trajectory, then removes this guidance before training so the student learns from clean demonstrations. On τ2‑bench, PROOF-Gen recovers 93% of failed scenarios, boosting Qwen3‑4B‑Instruct‑2507’s Pass^1 from 0.132 to 0.529 and improving Gemma 4 E4B‑it by 7.2pp on BFCL v4 multi‑turn, while also raising deployed on‑device model performance by up to 5.0pp across response‑quality metrics.
arXiv:2603.02041v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are predominantly trained on English-centric data, resulting in uneven performance for smaller languages. We stu...
arXiv:2608.24727v1 Announce Type: cross Abstract: EEG foundation models pretrained via self-supervised learning promise transferable representations, but their generalization remains limited, especia...
arXiv:2608.24597v1 Announce Type: cross Abstract: Electroencephalography (EEG) is a widely used window into human brain function, but most EEG models remain tied to a one-dataset-one-model supervised...
EXAM$^2$ is a new benchmark for audio understanding that covers six languages and multiple modalities—speech, sound, music, mixed-audio, and visual images—providing 5,667 multiple-choice questions, 22,614 image instances, and 135,684 multilingual translations. It evaluates large audio language models (LALMs) and multimodal large language models (LLMs), revealing significant gaps in multilingual and cross‑modal performance. The authors also introduce Gemma3n-EXAM$^2$, a lightweight fusion model that improves multilingual results by up to 12.4% and multimodal results by 21.7% over a strong baseline.
The paper introduces a two‑layer detector to prevent ‘slopsquatting’—the risk of local coding LLMs fabricating Python package names that adversaries can pre‑register on PyPI. The first layer checks PyPI for existence, while the second uses a Random Forest classifier on ten name‑and‑metadata features; an import reconciler resolves naming mismatches. Embedded in a LangGraph state machine, the system retries at escalating temperatures and falls back to stronger models, achieving hallucination‑free code in 76% of 300 curated prompts and recovering additional runs through intra‑ and cross‑model retries.
The paper introduces RobustTests, a framework that improves reinforcement learning for code generation by synthesizing test cases from faulty code and refining rewards with a dense, stepwise function. It uses validator agents and behavioral clustering to filter out invalid or redundant tests, and incorporates pass‑rate‑based rewards to counter hallucination noise. Experiments on CodeContests and LiveCodeBench show that fine‑tuning Qwen3‑32B with RobustTests yields a 3% absolute performance gain over baseline methods.
The paper presents a systematic study of scientific image synthesis, comparing pixel‑based generation and programmatic approaches. It introduces ImgCoder, a logic‑driven framework that follows an "understand‑plan‑code" workflow to enhance structural precision, and SciGenBench, a benchmark that evaluates images for information utility and logical validity. The authors find that pixel‑based models exhibit systematic failure modes and that fine‑tuning large multimodal models on rigorously verified synthetic images consistently improves downstream reasoning performance.
arXiv:2608.24093v1 Announce Type: cross Abstract: Self-supervised representation learning for 4D point cloud videos is challenging because annotations are costly and reconstruction-based pretraining...
Neurosymbolic Alignment couples a 7B clinical language model with a graph‑based physiological world model to score candidate responses using homeostatic constraints, multi‑hop plausibility, and drug‑interaction penalties. This training‑time framework drives iterative on‑policy updates and achieves a 90.8% Clinical Safety Score on the CSB benchmark, outperforming ORPO, GPT‑4 (5‑shot), and a self‑correction pipeline. Ablation studies show that the HGNN scoring and iterative training are the key contributors to the safety gains.
arXiv:2602. 02304v3 Announce Type: replace Abstract: Large-scale foundation models exhibit behavioral shifts when subjected to interventions such as scaling, fine-tuning, reinforcement learning with human feedback, or in-context learning.
arXiv:2608.23869v1 Announce Type: new Abstract: While state-of-the-art generative models produce high-fidelity 3D meshes, these outputs lack the physical properties required for interactive simulatio...
arXiv:2608.23903v1 Announce Type: new Abstract: Visual foundation models are commonly adapted under the assumption that the appearance of incoming data may change while the semantic meaning of the pr...
QiMeng-ChipV-RTL is a multi‑agent framework that tackles the challenges of generating Register‑Transfer Level (RTL) code for industrial IP‑level hardware design. By partitioning long design documents into short, localized tasks and using hierarchical planning, localized code generation, interface‑consistent merging, and AST‑guided debugging, it scales to complex specifications. Experiments on the RealBench benchmark show ChipV-RTL achieves a 45.0% pass rate, outperforming state‑of‑the‑art LLMs and agents which reach only 21.6%.
arXiv:2506.12006v4 Announce Type: replace-cross Abstract: The cross-Modality Domain Adaptation (crossMoDA) challenge series, initiated in 2021 in conjunction with the International Conference on Medi...
SketchGuard is a Byzantine‑robust decentralized federated learning method that separates neighbor screening from model aggregation by using a Count Sketch representation. The approach mitigates a vulnerability where an adaptive adversary can hide large perturbations in the sketch’s null space, by adopting a commit‑then‑sketch protocol that ensures the sketch seed is chosen only after model commitment. The authors prove convergence in both convex and non‑convex settings, demonstrate that SketchGuard achieves state‑of‑the‑art robustness against six attacks—including the adaptive null‑space attack—across various network topologies and data heterogeneity, while reducing per‑neighbor communication to a model‑dimension‑independent size.
The paper introduces a method for improving fine-grained, temporally aligned outputs in Speech Large Language Models (SpeechLLMs) by replacing absolute timestamps with relative timestamps, which reduces vocabulary size and enhances generalization. It proposes a hybrid fine‑tuning strategy that fully fine‑tunes the timestamp‑augmented embedding layer and language model head while applying LoRA to decoder layers, and introduces a masked timestamp training objective to prevent over‑reliance on ground‑truth timestamps. Experiments show significant gains in timestamp prediction accuracy without compromising transcription quality.
The paper introduces the "Compression Trinity," a unified framework that jointly applies sparsity, quantization, and low‑rank approximations to compress large language models. It presents several methods—MKOR, SLoPe, OPTIMA, PATCH, and SLiM—that leverage these three pillars to accelerate training, reduce memory bandwidth, and recover accuracy, achieving significant speedups and accuracy gains over existing techniques. The results demonstrate that combining all three compression strategies is essential for efficient, scalable, high‑performance LLM deployment.